仓库/Skill 库

从零开始在 PyTorch 中构建 decoder-only Transformer，涵盖分词、预训练、监督微调，以及基于核心张量操作的对齐。Build decoder-only Transformers from scratch in PyTorch, covering tokenization, pretraining, supervised fine-tuning, and alignment using core tensor operations.

全栈量化研究平台：因子分析、系统化策略回测、真实交易成本建模、滚动前推 ML 验证、投资组合风险分析与稳健性测试，基于美股日频数据。为可复现研究而构建——非实盘交易。Full-stack quantitative research platform for factor analysis, systematic strategy backtesting, realistic transaction-cost modeling, walk-forward ML validation, portfolio risk analytics, and robustness testing on daily U.S. equity data. Built for reproducible research—not live trading.

用于探索 LLM Security 新兴研究主题的研究路线图与文献综述仓库。A research roadmap and literature review repository for exploring emerging research topics in Large Language Model Security.

本地优先的 CLI，用于评估 MCP server 的开销与能力，数据不离开本机。A local-first CLI that measures what an MCP server costs and what it can do. Nothing leaves your machine.

探索用于测试 ML、LLM 和 Agentic AI 系统的 AI 安全工具、论文与框架Explore AI security tools, papers, and frameworks for testing ML, LLM, and agentic AI systems

80/20 Launch Audit：由资深开发者主导的上线审计，深入应用并证明问题所在。开源（MIT 协议）；付费托管审计 79 美元起。80/20 Launch Audit: a senior-dev launch audit that breaks into your app and proves what is broken. Open source (MIT); hosted audits from $79.

活文献综述：面向网络安全领域的现代数据架构Living literature review: Modern data architecture for cybersecurity

通过硬件 IP 核在硅层级抑制 LLM 幻觉，强制 LLM 输出的语义完整性Enforce semantic integrity in LLM outputs with a hardware IP core that suppresses hallucinations at the silicon level

为将 MCP 服务器推向企业部署的团队提供的实用 MCP 就绪、安全与无状态迁移资源。Practical MCP readiness, security and stateless-migration resources for teams moving MCP servers toward enterprise deployment.

基于 Semantic Scholar 与 Crossref 的安全加固、双模式学术搜索工作台。A security-hardened, dual-mode scholarly search workbench powered by Semantic Scholar and Crossref.

Agent 优先、安全优先的工具包，用于开发和定制 reMarkable Paper Pure 平板。Agent-first, security-conscious toolkit for developing and customizing reMarkable Paper Pure tablets

分析葡萄牙公共养老金体系历史融资的可复现研究包Reproducible research package analyzing the historical financing of Portugal's public pension system

纯 Go 实现的分布式列式 SQL 分析引擎——对 S3 上 Parquet 进行向量化执行，无需 JVM、无需 CGo。以 Trino 级别的能力提供相当性能，却只需极小资源占用。原生支持 IPv4/CIDR/MAC 等网络类型，适用于遥测和安全场景。Distributed columnar SQL analytics engine in pure Go — vectorized execution over Parquet on S3, no JVM, no CGo. Trino-class performance on a fraction of the footprint. Network-native IPv4/CIDR/MAC types for telemetry and security workloads.

即时评估课程资格，输出明确的通过/未通过结果以及定制化的入学测试。Instantly evaluate course eligibility with clear pass/fail results and tailored entry assessments.

面向 20–30 岁高血压与糖尿病高风险人群的 AI 医疗 — 教练关联的饮食与运动指导平台AI healthcare for 20–30s at risk of hypertension & diabetes — Trainer-Linked Diet & Exercise Coaching Platform

AI 助手的安检守卫。检查 AI 尝试调用的每一个工具，防止数据泄露与未授权命令。A security guard for AI assistants. It checks every tool an AI tries to use, preventing data leaks and unauthorized commands.

面向 Linux VPS 服务器的自主式 AI 驱动 SOC 安全 agent 与实时威胁分类器。包含针对 48 种攻击类别微调的 DistilBERT Transformer、DDoS 异常检测器、系统资源监控器、带 HTML 邮件告警的中央告警引擎，以及 REST 健康诊断接口。Autonomous AI-powered SOC Security Agent & Real-Time Threat Classifier for Linux VPS servers. Features a fine-tuned DistilBERT transformer (48 attack classes), DDoS anomaly detector, system resource monitor , central alert engine with HTML email reporting, and REST health diagnostics.

加密货币/数字资产研究，包含链上分析、DeFi 流动性、协议风险以及可复现研究工具。crypto/digital-assets research, on-chain analytics, DeFi liquidity, protocol risk, and reproducible research tooling.

自包含的系统综述工作台——多库检索、双评审筛选、AI 辅助批判性评价（RoB 2、ROBINS-I、GRADE…）。单 HTML 文件，无后端，数据不离开浏览器。Self-contained systematic review workbench — multi-database search, dual-reviewer screening, AI-assisted critical appraisal (RoB 2, ROBINS-I, GRADE…). One HTML file, no backend, data never leaves your browser.

基于人类基准的 LLM 生成 Infrastructure-as-Code 安全基准测试。100 场景 × 12 模型配置 = 1,196 个工件，由 Checkov/Trivy/KICS 扫描，并与 634 个人工编写模板对比。所有模型的漏洞密度均为人工的 3.2–3.9 倍。arXiv:2608.28021Human-anchored security benchmark for LLM-generated Infrastructure-as-Code. 100 scenarios × 12 model configs = 1,196 artifacts scanned by Checkov/Trivy/KICS, compared against 634 human-written templates. Every model: 3.2–3.9× human vulnerability density. arXiv:2608.28021

论文《Airflow Representation in Hygrothermal Modelling for Moisture Risk Assessment of Building Envelopes: A Critical Review of 45 years》（A. Sarkany 等）数据集。Dataset for "Airflow Representation in Hygrothermal Modelling for Moisture Risk Assessment of Building Envelopes: A Critical Review of 45 years" by A. Sarkany et. al.

人工智能与软件开发公司；构建智能系统、自动化工具、定制软件及开源技术。شرکت هوش مصنوعی و توسعه نرم‌افزار؛ سازنده سیستم‌های هوشمند، ابزارهای اتوماسیون، نرم‌افزارهای سفارشی و فناوری‌های متن‌باز.

10 门 AIMarket 学院实战课程 —— 编排、oracles、MCP 安全、Agent 经济（en/ru/es/fr/zh）。10 hands-on AIMarket academy courses — orchestration, oracles, MCP security, agent economy (en/ru/es/fr/zh).

🌐 利用多跳图推理结合知识图谱和向量搜索，挖掘隐藏的全球供应链风险，获得更深层洞察。🌐 Map hidden global supply chain risks with multi-hop graph reasoning combining knowledge graphs and vector search for deeper insights.

付费 MCP 服务器：以通俗英语确定性解码任意 Base 主网交易，输出严格 JSON，包含资产转移、对手方、风险标志、以美元计价的 gas。通过 x402 计费，每次调用 $0.02。Paid MCP server: plain-English deterministic decode of any Base mainnet transaction. Strict JSON with assets moved, counterparties, risk flags, gas in USD. $0.02/call via x402.

AI 工程作品集，聚焦可控 Agent、可复现研究与本地生成式 AI 基础设施AI engineering portfolio focused on controllable agents, reproducible research, and local generative AI infrastructure.